Intelligent Lightning Warning System

Transcription

1 Intelligent Lightning Warning System Geeth Jayendra, Rohan Lucas, Sisil Kumarawadu, Lilantha Neelawala, Chathura Jeevantha, P. Dharmapriya Department of Electrical Engineering, Faculty of Engineering, University of Moratuwa, Sri Lanka. {geeth, lucas, sisil,lilantha_02, chatura_02, Abstract In developing countries like Sri Lanka, lightning warning systems are rare, and simple low cost methods adopted have not lived up to expectations. It would thus be very advantageous if warnings could be made localized to a specific area. This paper presents a method that is affordable to the ordinary person and informs him of the risk of lightning hazards in advance. The system is simple, yet effective, and produces specific warnings accurately using an effective technology. The lightning threat is assessed by monitoring the static electric field between the thundercloud and the earth using a usermade made field mill unit. This triggers an alarm, or a rotating beacon according to the degree of the lightning threat, with the aid of a simple neural network. Intra cloud flashes, which occur in the vicinity of imminent lightning flashes in the region, emits radio frequency signals. By combining this as an additional input to the neural network, more accurate predictions can be made. Test results are also presented on the prototype lightning warning system developed at the University of Moratuwa. 1. INTRODUCTION Lightning is a formidable natural phenomenon that is responsible for a considerable number of deaths in Sri Lanka. The number of injured people in lightning incidents, as reported, is generally over 50 in each year [1]. Since there is a lack of reported information, actual numbers can be much higher. Damage to property and loss of life by lightning flashes are on the increase. This is due to the increase of the population and the use of modern electrical and electronic equipment in our day-to-day lives, without proper regard to safety precautions. Lightning activity over Sri Lanka shows peaks during the two Inter-monsoon seasons of March-April and October- November [2]. During these periods, convective clouds (cumulonimbus) develop over most parts of the island, mostly during the afternoon or evening. The well-developed cumulonimbus clouds produce thunderstorms. Since thunderstorms develop under many atmospheric conditions, it would be advisable to be alert and launch precautionary steps to reduce lightning hazards during all seasons. It is important to note that damages are not only caused by direct lightning strikes, but also from indirect strikes [3]. Surges in power distribution lines caused by lightning may reach households through the utility supply and destroy valuable electronic equipment instantly. Loss of lives due to lightning incidents is mainly due to the fact that precautionary steps have not been taken to reduce or avoid lightning hazards [4]. Some precautionary methods of reducing consequences are disconnecting electrical/electronic equipments from the main power supply, disconnecting antennas from television sets, avoiding handling of electrical/electronic equipment, avoiding touching or standing close to metal structures and limiting/avoiding the use of telephones, when the possibility of a lightning risk is high [5]. It can be concluded that by giving early warnings of imminent lightning threats, the damages due to lightning can be mitigated. It would be invaluable if one can be updated with the risk of lightning threat in the vicinity, so that adequate precautions can be taken when the threat increases above a threshold level. The lightning warning system developed (Fig. 1) is a hybrid design which uses a combination of sensing the static electric field and the radio frequency discharges from intra-cloud lightning discharges in the vicinity. Field Mill RF Detector Processing Unit Fig. 1 Lightning Warning System Indicator Alarm This paper is organized as follows. Section 2 outlines the electrical characteristics of thunderclouds, and the detection of lightning is given in section 3. Acquisition of the input information and the subsequent use of a neural network to produce a warning signal is described in section 4. Section 5 describes the test results and section 6 concludes the paper. 2. THUNDER CLOUDS AND LIGHTNING The basic composition of the cloud charge structure includes a net positive charge near the top, a net negative charge below it and an additional positive charge at the bottom of the cloud [6] as shown in the simple model in Fig. 2. Lightning is an electrical discharge associated with charged cumulonimbus clouds. They are usually classified as Cloud to Ground Flash (ground flash), Cloud Flash (intra cloud flash), and Air Flash (discharge between clouds and the atmosphere) /07/$ IEEE 19 ICIAFS07

2 The ground flash, although not the most prevalent, is the most important as a hazardous event. It is the electrical discharge between the negative charge centre of a cloud and the positive charge developed on the ground. Under certain conditions, positive ground flashes may also occur between the positive charge centre of a cloud and the earth. 12 km 7 km km 38c 40c 2c Fig. 2. Gross charge structure of thundercloud When lightning flashes occur, sudden electric field changes are observed [7]. This is due to the charge removal from the cloud (Fig. 3) In order to measure the field, a field mill is used [10]. In the field-mill, a grounded and segmented top plate rotates so as to cover and uncover a fixed, similar field detecting plate beneath it [11]. The importance of the field-mill method is that it can help produce early warning, sufficiently in advance, if the output it produces is correctly interpreted. Other methods such as RF detection, magnetic field change measurements, and sudden electric field change measurements require lightning flashes to actually occur. Thus, the most suitable method would be the monitoring of the static electric field between the thundercloud and the earth as described in section 2. It is clear that by observing the electric field strength, a prior warning can be delivered. As the thundercloud is approaching a particular location, the static electric field measured at that location increases. The measured field would gradually decrease as the cloud moves away. Therefore, the measured strength of the static electric field can be used to quantify the threat. Here, the vertical component of the electric field can be considered. Fig. 3. Sudden electric field changes 3. DETECTION OF LIGHTNING The lightning flash is a huge current flow which occurs within a very short time period of the order of microseconds. Thus, it generates radio frequencies (RF), and rapid changes in electric fields and the magnetic fields [8]. A. Radio Frequency Measurement RF detectors measure energy discharges from lightning. They can determine the approximate distance and direction of the threat. Lightning emits electromagnetic energy in the frequency range from below 1 Hz to 300 MHz. Lightning also produces noise like bursts of electromagnetic radiation lasting for tens to hundreds of microseconds. This can also be used in a method to detect lightning. In our work, RF interruptions are detected using the IC CXA1619BS [9]. The circuit is initially tuned to 80 MHz. The output produced in response to a lightning flash is fed to the microcontroller using an opto-coupler. The interrupt pin of the microcontroller is used for receiving the signal. B. Field-Mill to Measure Electric Field Electric field measurement is a vital part of the product. The aim is to measure the static electric field produced by the thundercloud and to detect the electric field changes due to lightning. Fig. 4. Sensor plate exposed to the electric field Fig. 5. Sensor plate is shielded from the electric field The electric field mill is a device based on electrostatic induction principle. It consists of two electrodes. One electrode is constructed as a set of rotating vanes. The other electrode is the sense plate, which is only periodically exposed (Fig.4) to the field due to the rotation of the rotating vanes (Fig. 5). 20

3 The rotating vanes are located over the sensor plate so that it periodically shields and exposes the sensor plate to the electric field. To do this properly, the rotor must be grounded. The sensor plate is grounded through an amplifier, which converts the sensor plate's ground current to a voltage. As the sensor plate is exposed to the field, the field induces ground currents as it attracts or repels charge from the sense plate. When the plate is shielded from the field, the induced charge drains away. This induces an alternating ground current that is proportional to the electric field strength. These raw signals are generally weaker and should be amplified before feeding the processing unit. The magnitude of the voltage output by the amplifier is used as a measure of the magnitude of the static electric field between the thundercloud and the earth. This type of a field-mill unit can be usermade and installed in places where heavy lightning activities are experienced allowing people to know the threat in advance and take suitable precautions to avoid potential damage. C. Measurement of Static Field AC current input from sensor plate A/D converter of microcontroller current to voltage converter Rectifier Op Amp Low pass filter (Active filter) Fig. 6. Procedure of detecting the static field The alternating ground current, which is induced on the sense plate, is first converted to a voltage signal by the current-tovoltage converter. This voltage signal is then amplified by the operational amplifier and sent through the low pass filter. This signal is rectified to produce a dc voltage between 0 to 5 V, and supplied to the analog-to-digital converter of the microcontroller. This dc voltage level varies according to the static electric field that the field-mill is exposed to. The relation of motor speed to frequency of induced AC ground current can be described as follows. The set time for the basic step angle is 20 ms. Since the motor has a basic step angle of 7.5 o, the time taken for one complete rotation can be obtained as 20 X 360/7.5 = 960 ms. Thus, the frequency of the rotation is Hz. Since the stator has four openings in the field-mill and the utilization of a 4 vane shutter, the effective frequency of exposing and covering of the sense plate from the electric field is obtained as X 4 = 4.17 Hz. This is the frequency of induced ground current. The relation is verified by the amplified signal obtained from the sensing plate (Fig. 7). The signal has a period of 240 ms corresponding to Hz. Fig.7. Frequency of AC ground current 4. USE OF NEURAL NETWORK A. Neural Network A Neural network (NN) is trained to learn the input-output relationship to be able to produce the threat level real-time. The two inputs are the radio frequency and the static field. The single output is the warning signal or the threat level. The desired outputs for the training of the NN are not available at the time it produces the actual output (the warning signal). The desired output is obtained based on the number of radio frequency interferences detected during the period of 10 minutes after the warning signal is produced. Therefore, the NN is trained based on the error signal obtained by comparing the actual output of the NN and a desired output that is obtained by detecting the RF interferences over a period of 10 minutes after the actual input is produced. The NN in Fig. 8 is trained using a re-enforcement like learning approach. The perceptron learning algorithm [12] that is used for self learning is explained. Fig.8. Neural Network for the unit X 0 = Bias input X 1 = Radio interruptions X 2 = Electric field W 0,1,W 0,2,,W 2,3 = Connecting weights (see Fig. 10) 21

4 Start 1 Minute Start Motor Measure Electric t = G(RF1) α w= w+ η[ F t]x ( w x) α + 1 α 2 Fig. 9.Classes of the Output Stop Motor No Y 3 = 1 Yes 1 Minute Stop 10 Y 2 = 1 Yes No Average Threat Fig. 10 Weights relevant to the perceptron Outputs of the neural networks were classified into three main classes (Fig. 9). C1 class was used to represent High-threat warning, C2 for Average-threat warning, and C3 for Nothreat. Y 1 = 1 Yes High Threat Stop No B. Perceptron learning algorithm Figure 11 shows the flow chart of the system process. Y 1, Y 2 and Y 3 are the outputs from the neural network for High threat, Medium threat and No threat status. Times T 1 and T 2 are 10 minutes and 1 minute, respectively. Function G is used to obtain the target output. Initially, the output Y 3 is checked from the first perceptron and if it is one, the motor is stopped with a wait time of 10 minutes. After 10 minutes, the motor starts again and the outputs are checked. If Y 3 = 1, the same procedure is followed. If Y 3 = 0, the output of Y 2 is checked. If Y 2 is zero, the Average-threat warning is given and the motor is stopped with a wait time of one minute. The function G calculates the target values, t, or the desired outputs for the NN based on the signals from the radio frequency unit. The weights are then updated accordingly. Then, the motor starts and the loop also starts from the beginning. In the loop, if Y 2 is also zero, go to the next step, where the output Y 3 is examined. If Y 1 = 0, issue an Average-threat warning and wait one minute after stopping the motor. Then, the function G updates its weights based on the radio frequency interruptions detected. The motor starts again and so does the loop. If Y 1 = 1, issue the High-threat warning. Then, stop the motor and wait for 10 minutes. Here, also the function G updates its weights, and after another 10 minutes the motor and loop are started from the beginning. 10 t = G(RF1) α w= w+ η [ F t]x ( w x) α + 1 α 2 Fig.11. Flow Chart of the System Process 5. TESTING The field-mill was tested in the laboratory environment using an artificial field, which was generated by applying a DC voltage between two parallel plates. The strength of the field was adjusted to be similar to the actual electric field existing before lightning strikes. The potential of lightning was mainly categorized as low, medium and high. The voltage induced in the sensing plate of the field-mill is varies according to the strength of this static field. Lightning strikes are simulated by sparking of sphere gaps in the High-voltage laboratory of the University of Moratuwa. The following oscilloscope images (Fig. 12, 13 & 14) depict the raw signal induced in the sense plate and after initial filtering of those signals for field strength of 1 kv/m, 2 kv/m and 3 kv/m, respectively. 22

5 Fig.12 Amplified signal from sense plate, before filtering Fig. 13. Field Strength 1kV/m DC Output from Feild Mill(V) Electric Field (kv/m) Fig.16. Field mill output vs. Electric field The DC output from the field-mill was plotted against the applied artificial electric field and those results were shown in the Fig.16. Fig. 17 shows the test result when the RF detector detects the lightning discharge. In the lightning warning system is shown in Fig. 18. Three output levels, namely, no threat, average threat, and high threat, respectively are indicated by a green light, amber light, and a red light and a siren allowing the user can take precautionary actions accordingly. Fig. 14 Field Strength 2kV/m Fig. 17 Output of the RF detector Fig. 15 shows a DC voltage derived from the signal from the sensing plate, which is fed to the analog to digital converter of the microcontroller. Fig. 15 Field Strength 2kV/m The CXA1619BS IC used in the RF receiving circuit costs 2 US$ and the total cost of the PICs used for motor control and NN implementation is less than 10 US$. Thus the product is relatively cheap to make. Fig.18. The Product 23

6 6. CONCLUSIONS A simple low-cost lightning warning system, based on static electric field caused by the clouds and the number of radio frequency interruptions are utilized to give lightning warnings, has been presented in this paper. A simple neural network has been utilised to make the warning decisions. The measurements of the static electric field between the thundercloud and the earth are obtained using a field-mill. The radio frequency interruptions are detected using a normal radio receiver circuit. The system has been thoroughly tested under laboratory conditions and has shown to be reliable. Test results validate the prototype and will be useful for future developers to come up with a durable and affordable solution for the damage caused by lightning. Further testing would be required to finetune under live situations. [10] Natalie Murray, Philip Krider, James Dye, Surface observations of the electric field and the radar reflectivity of decaying thunderstorm anvils and debris clouds at the nasa kennedy space center, Journal of applied meteorology and climatology, American Meteorological Society, November [11] J. Monta, J. Bergas, B. Hermoso, Natural electric field measurements applied as a predictive system against lightning in power systems, Proceeding (369) Power and Energy Systems, ACTA press, Canada, [12] J. Ma, The Object Perceptron Learning Algorithm on Generalised Hopfield Networks for Associative Memory, Neural Computing & Applications, vol. 8, pp 25-32, March ACKNOWLEDGEMENT The authors acknowledge the contributions of Dr. Fernando of the University of Colombo and Mr. Abhayasinghe Bandara of the Meteorological Department of Sri Lanka. REFERENCES [1] Chandima gomas, Richard Keithel, Munir ahamad, Developing a lighting awareness program model for third world based on American South Asian countries, National Lightning Safety Institute, March [2] Nadia Fazlulhaq, Inter-monsoon season brings thunderstorms and mini cyclones,the Sunday times online, vol. 41, pp 42, Mar [3] C. Gomes, M. Ahmed, F.H. Shuvra, and K.R. Abeysinghe, "Lightning accidents and awareness in South Asian-Experience of Sri Lanka and Bangladesh" International Conference on Lightning Protection, Kanazawa, Japan, 2006 [4] Arun Kulshreshtha, Protecting developing countries from the dangers of lightning, Science news, June [5] Rod Brouhard,, How To Avoid Lightning Strikes, Your Guide to First Aid,May [6] R.B. Anderson, R.H. Golde (edited by), Lightning, Volume 1, Physics of Lightning, 1977, ch. 13 [7] Vladimir Rakov,Martin Uman, Lightning Physics and Effects,Cambridge University Press, 2003, ch.3, ch. 17 [8] Martin Uman, Comparison of the RF Frequency Spectra of hemp and Lightning, Storming Media, March 1991 [9] CXA1619BS Technical Documentation, Sony Corporation, Japan. Available at: 24